Land cover classification at three different levels of detail from optical and radar sentinel sar data: A case study in cundinamarca (colombia)

Abstract

In this paper, the potential of Sentinel-1A and Sentinel-2A satellite images for land cover mapping is evaluated at three levels of spatial detail; exploratory, reconnaissance, and semi-detailed. To do so, two different image classification approaches are compared: (i) a traditional pixel-wise approach; and (ii) an object–oriented approach. In both cases, the classification task was conducted using the “RandomForest” algorithm. The case study was also intended to identify a set of radar channels, optical bands, and indices that are relevant for classification. The thematic accuracy of the classifications displays the best results for the object-oriented approach to exploratory and recognition levels. The results show that the integration of multispectral and radar data as explanatory variables for classification provides better results than the use of a single data source.